HRS/HCAP Algorithm in HRS Core

Rich Jones, Doug Tommet

Original presentation date: 19 Dec 2024

2025-04-13

Contents

  1. Goal of project

  2. Analysis plan and progress

  3. Some selected results

  4. Next steps and timeline

Project Goal

  • The goal is to determine to what degree the Core data can be used to approximate the HRS/HCAP classification.

  • The procedure will be to classify 2016 HRS/HCAP participants into cognitive impairment categories (normal, mild cognitive impairment, dementia) using only data from the HRS 2016 Core interview (i.e., omitting all HRS/HCAP data).

  • This is only one of several possible approaches to cognitive classification of Core participants.

Figure 1: HRS algorithm
Statistical analysis plan GDoc
Detailed analysis results HTML
Mplus output final factor model TXT

Progress

  • 1. Factor model for general cognition in HRS/Core (in R)

  • 2. Adjustment, standardization and normalization (in R)

  • 3. Equivalent cut point for identifying severe and moderate cognitive impairment to that used in HRS/HCAP algorithm

  • 4. Operationalize functional impairment in HRS/Core

  • 5. Run and compare algorithms

  • 6. Reporting

Selected results

Single factor model for HRS/Core cognition

Table 1:

Model 2: Fit statistics
Fit statistic Value
RMSEA : Estimate 0.057
CFI 0.961
SRMR 0.041
Table 2:

Model 2: Factor loadings
Item Label Std Factor Loading
vdori Orientation to time 0.485
vdlfl1z Animal naming 0.591
vdlfl2 Scissors & cactus 0.678
vdlfl3 President & vice-president 0.599
vdwddelz Delayed word recall 0.589
vdexf7z Number series 0.642
vdsevens Serial sevens 0.706
vdcount Count backwards from 20 0.567

Note: items with a “z” at the end are continuous indicators, the others are categorical indicators.

  • The single factor model fits well by conventional fit criteria.

  • Like the HRS/HCAP factor models, we do not include immediate memory performance in the model.

  • We do include naming and serial 7s, which are not in HRS/HCAP models

  • During the meeting we discussed that the model may be too heavy in “other” (notably serial 7s, number series) and less “memory” and this may be problematic for AD. We will explore other solutions.

Figure 2: Distribution of factor scores by HCAP normative reference status

In the normative
reference group

NOT in the normative
reference group

Figure 3: Distribution of factor scores by HCAP normative reference status, including HRS sample

In the normative
reference group

NOT in the normative
reference group

NOT in HRS/HCAP

Figure 4: Cognitive factor score in HCAP vs HRS

Figure 5: Bland-Altman plot of cognitive factor score in HCAP vs HRS

Figure 6: Distribution of HCAP cognitive factor score by number of domains impaired

Figure 7: Distribution of HRS cognitive factor score by number of domains impaired
Table 3:

Distribution of the number of impaired domains in HCAP
Characteristic N = 2,9931
Number of domains impaired
    No domains 1,935 (65%)
    1 domain 587 (20%)
    2+ domains 471 (16%)
1 n (%)

Number of domains impaired, based
on full HRS/HCAP algorithm using 5
individual domains

quantile(hcap$TF, probs = c(.15, .35), na.rm=TRUE)
     15%      35% 
36.43890 43.77599

Centiles of adjusted, normalized, standardized factor score in HRS/Core

Table 4:

Crosstab of number of domains impaired (HCAP) vs matched factor scores (HRS)
Level of cognitive impairment (HRS)
 Total
None Mild Severe Unknown
Number of domains impaired




    No domains 1,528 307 88 12 1,935
    1 domain 312 164 99 12 587
    2+ domains 101 101 218 51 471
Total 1,941 572 405 75 2,993

Number of domains impaired,
based on full HRS/HCAP
algorithm using 5 individual
domains

Based on matching centiles of adjusted,
normalized, standardized factor score in
HRS/Core

Progress

  • 1. Factor model for general cognition in HRS/Core (in R)

  • 2. Adjustment, standardization and normalization (in R)

  • 3. Equivalent cut point for identifying severe and moderate cognitive impairment to that used in HRS/HCAP algorithm

  • 4. Operationalize functional impairment in HRS/Core

  • 5. Run and compare algorithms

  • 6. Reporting

Figure 8: HRS algorithm